Data for "When to Revoke Completion: Evidence-Gated Completion Records for Multi-Step Robot Manipulation under Unreliable Goal Checks"
收藏官方服务:
资源简介:
Trial records, model-based predictions, vision-language-model calibration data and original study registrations for the paper "When to Revoke Completion: Evidence-Gated Completion Records for Multi-Step Robot Manipulation under Unreliable Goal Checks". The archive contains four pre-registered simulation studies (Studies C-F, 3,040 trials in a Unity/PhysX pick-and-place scene), the per-trial event logs and sealed summaries, the predictions generated before each run, the camera frames and VLM answers of the goal-check calibration, and the frozen source files with the SHA-256 registrations they were checked against. All trials were run in simulation. See DATA_README.md for the layout. Code: https://github.com/hlf885/evidence-gated-revocation
提供机构:
Zenodo创建时间:
2026-11-04



